Quantifying intersection performance is often a key aspect of transportation engineering studies. These studies commonly use microsimulation tools to evaluate intersection performance for vehicle delay by averaging results from several simulation runs (each with a different pseudorandom number seed) for a set of traffic and control conditions. Previous research has showed that this approach may lead to a bias in the estimation of both the mean and the standard deviation of field delay. This paper examines the issue of explicitly modeling the day-to-day variability in intersection delay by using microsimulation modeling. The results show that the prevalent method of using multiple runs, each with the same traffic demands but with different pseudorandom number seeds, does not adequately capture the day-to-day variability observed in the field. Consequently, two alternative methods for modeling day-to-day variability of intersection performance are presented and examined.